From the talk, The bottleneck moved
Every screen has five states. The tool builds one.
Your users live in the other four. These are the three files I use every day to make an AI builder draw them, audit its own screens, and stay on brand. They are yours.
five-states-prompt.md
Run it after every build prompt, one screen at a time. It produces the empty, loading, partial, and error states the tool skipped, then lists what it still has not handled.
screen-audit-prompt.md
Ten questions the tool answers about its own output before you accept a screen. Question nine is the whole method: what did you decide that I never asked for?
BRAND.md
One file that lives in your repo: your one sentence, your nouns, your voice rules, your tokens. Every session starts on brand instead of on default.